Hoarseness is a common symptom of multiple laryngeal diseases such as inflammation, paralysis, neurologic disease, or laryngeal cancer. Many patients with these diseases are not diagnosed with the correct underlying cause of the hoarseness early enough. Therefore, healthcare providers need better methods to screen for and evaluate different types of hoarseness. Currently, a combination of tools are used to evaluate voice disorders in specialty clinics such as patient history, perceptual voice evaluation, and laryngoscopy. We want to better understand how providers with different medical backgrounds evaluate patients with voice complaints. We are most interested in seeing how history, perceptual voice evaluation, and laryngoscopy impact decision-making and diagnosis. In addition, our group has developed a machine learning algorithm that analyzes voice to detect the presence or absence of a laryngeal mass. We want to see if this algorithm could be clinically useful for generalist providers. To address these questions, a group of clinician evaluators including general practitioners, otolaryngologists, and speech language pathologists, will be recruited remotely. Subjects will be asked to complete an electronic questionnaire with patient case scenarios, asking them to evaluate hoarse voice samples and laryngoscopy exams, with and without case history. For perceptual voice sample evaluations, clinician performance will be compared to the algorithm’s classification of whether a hoarse voice is from someone with a laryngeal mass. From there we will see if clinician detection of laryngeal masses from voice could be improved with this algorithm. If the algorithm has better performance than clinicians, then it may be clinically useful as a screening tool in the future. Our results will help us understand how evaluations for hoarseness are done and can be improved.

Hoarseness is a common symptom reported to generalist healthcare providers, with approximately 1% of the clinical population being affected by it each year. It can be caused by multiple etiologies, such as hoarseness due to a cold, acid reflux, or laryngeal cancer. Perceptual evaluation of the voice is inaccurate, and it is therefore difficult to differentiate between hoarseness requiring urgent referral for specialty evaluation (i.e. laryngeal cancer) versus a disorder that could be managed without specialty care (i.e. acute laryngitis). The current gold standard of diagnosis for hoarseness is laryngoscopy, an in-clinic endoscopy recording of the larynx performed by an otolaryngologist specialist. Our research team seeks to improve perceptual voice evaluation by developing and testing machine learning algorithms which analyze voice for underlying pathology, beginning with an algorithm which screens voice for laryngeal masses. We hypothesize that our algorithm will have greater than 80% sensitivity and specificity in the classification of voice samples from patients with laryngeal masses. To test this, we are developing a large, prospective database of voice samples from a laryngology clinic using a smartphone application. Subjects are adult patients presenting to the laryngology clinic, with and without voice disorders, who have had a recent laryngoscopy exam and no laryngeal surgery within the past three months. We are collecting patient history which could influence voice quality, such as age, gender, alcohol use, smoking history, and subject-perceived voice disorder impact. After collection of the voice sample and patient history, cases are classified into underlying pathologic categories. We see recruitment of a well-classified and prospective patient population with a range of voice disorders. This work could lead to improved screening of patients with hoarseness in underserved and primary care settings, and more appropriate and timelier specialist referrals and treatment.